This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

Volt-VAr Control and Energy Storage Device

Operation to Improve the Electric Vehicle Charging
Coordination in Unbalanced Distribution Networks
Carlos F. Sabillon-Antunez, Ozy D. Melgar-Dominguez, John F. Franco, Member, IEEE,
Marina Lavorato, Member, IEEE, and Marcos J. Rider, Senior Member, IEEE

Abstract—In this paper, a new approach is presented to N OMENCLATURE

solve the electric vehicle charging coordination (EVCC) problem
considering Volt-VAr control, energy storage device (ESD) oper- Constants:
ation and dispatchable distributed generation (DG) available in G
ζn,t Energy cost at node n at time t
three-phase unbalanced electrical distribution networks (EDNs). EV SD
Dynamic scheduling for the EVCC is proposed through a
ζc , ζc EV and ESD energy curtailment costs
step-by-step methodology, which solves a mixed integer linear ζ SD Unit cost for the active power taken from the ESD
programming (MILP) problem for the whole time period. The αn,f , βn,f Load voltage dependence parameters of the active
objective is to minimize the total cost of energy purchased from and reactive power at node n, phasef
the substation and DG units, the cost of energy curtailment γ(e, m, f ) Function that indicates whether EV e is connected
on electric vehicles, the cost of energy injected from the ESDs
and the cost of energy curtailment on the ESDs. The Volt-
at node m and phase f
VAr control considers the management of on-load tap changers, γ(u, m, f ) Function that indicates whether ESD u is connected
voltage regulators, and switchable capacitors installed along the at node m and phase f
grid. Furthermore, the formulation takes into account the voltage ηe Charging efficiency of the charging station for EV e
dependency of the loads, while the steady-state operation of ηu+ , ηu− Charging and discharging efficiencies of the charging
the unbalanced distribution systems is modeled using linear
constraints. The proposed model was tested in a 178-node three
station for ESD u
phase unbalanced EDN considering a one-day time period. ∆t Duration of time interval t
∆cb,∆vr Maximum number of operations allowable over the
Index Terms—Electric vehicle charging coordination problem,
energy storage devices, mixed integer linear programming, Volt- time period for the CBs and VRs
VAr control, voltage dependent load model. φn Minimum power factor for the operation of the DG
at node n
Bn Maximum number of CB modules at node n
ACRONYMS sh
Bmn,f Shunt susceptance of circuit mn, phase f
CB Capacitor bank EV SD
E e , E u Energy capacity of EV e and ESD u
DG Distributed generation
Eeini Initial state of charge of EV e
DoD Depth of discharge SD
Eu f Goal for the state of charge of ESD u at the end of
EDN Electrical distribution network
the time period
ESD Energy storage device
EuSDi Initial state of charge of ESD u
EV Electric vehicle
I mn Maximum current flow magnitude of circuit mn
EVCC Electric vehicle charging coordination im∗ re∗
Imn,f,t , Imn,f,t Imaginary and real part of the estimated current
MILP Mixed integer linear programming
of the VR in circuit mn, phase f , time t
MINLP Mixed integer nonlinear programming o
Pn,f,t , Qon,f,t Nominal active and reactive power demand at
OLTC On-load tap changer
node n, phasef , time t
SOC Initial state of charge EV ch
VR Voltage regulator Pe Maximum power consumption of EV e
SD
VVC Volt VAr control Pu Maximum charging/discharging active power of ESD
u
This work was supported by the Brazilian institutions CAPES, CNPq and G
FAPESP. Pn Maximum active power of the DG at node n
C. F. Sabillon and O. D. Melgar are with the Department of Electrical
Engineering, UNESP – São Paulo State University, Ilha Solteira 15385-000,
Qesp
n CB module reactive power capacity at node n.
G
São Paulo, Brazil. (E-mails: {cfsa27, ozzy.damedo}@gmail.com). Qn,QGn
Maximum and minimum reactive powers of the DG
J. F. Franco is with the UNESP – São Paulo State University, Rosana at node n
19273-000, São Paulo, Brazil (e-mail: j.f.franco@ieee.org).
M. Lavorato is with the Electrical Engineering Faculty, PUC/CAMPINAS – Rmn,f,h Resistance of circuit mn between phases f and h
Pontifical Catholic University of Campinas, Campinas 13086-900, São Paulo, %Rmn Regulation percentage of the VR in circuit mn
Brazil (e-mail: marina.oliveira@puc-campinas.edu.br). T pmn Maximum number of steps of the VR in circuit mn
M. J. Rider is with the Department of Systems and Energy, UNICAMP
– University of Campinas, Campinas 13083-852, São Paulo, Brazil (e-mail: V ,V Upper and lower voltage magnitude limits
mjrider@dsee.fee.unicamp.br). Vo Nominal voltage

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

∗ ∗
im re
Vn,f,t , Vn,f,t Imaginary and real part of the estimated voltage as input for the VVC. A multi-objective optimization for
at node n, phase f , time t the VVC was presented in [2]. In this work, a strategy was
Xmn,f,h Reactance of circuit mn between phases f and h formulated based on an hourly load forecast for the next
Variables: day, taking into account the active power demand reduction
Bn,t Modules of the CB in operation at node n, time t and the voltage magnitude deviation. Finally, a deterministic
btmn,f,t,k Binary variable that describes the tap position of the framework based on a mixed-integer quadratically constrained
VR in circuit mn, phase f , time t, position k programming problem capable of optimally controlling the
EeEV Energy of EV e at the end of the time period Volt-VAr problem was formulated in [3].
EeEV c Energy curtailment on EV e at the end of the time In recent years, energy storage devices (ESDs) have
period emerged as a solution to even out the power mismatch between
SD
Eu,T Energy of ESD u at the end of the time period renewable power generators and consumption. The ESDs store
EuSDc Energy curtailment on ESD u at the end of the time the surplus of power for use during time periods of low power
period generation. Nevertheless, the performance of ESDs has also
c(re) proven advantageous in improving the economic and technical
Imn,f,t,k Real part of the VR current calculation auxiliary
operation of the EDN [9]– [11]. Reference [9] discussed how
variable in circuit mn, phase f , time t, position k
cbim cbre the proper functioning of the ESDs in the EDN can improve
In,t , In,t Imaginary and real part of the current generated
the economic use of the existing generation, transmission
by the CB at node n, time t
Dim Dre and distribution infrastructure. A mixed-integer second-order
In,f,t , In,f,t Imaginary and real part of the load current at node
cone programming model was presented in [10] to solve the
n, phase f , time t
EV im EV re optimal operation problem of ESDs in a radial EDN. The
Ie,t , Ie,t Imaginary and real part of the current de-
presented approach was tested in two single-phase systems and
manded by EV e, time t
Gim Gre no dynamic control was presented for ESD charging. In [11],
In,f,t , In,f,t Imaginary and real part of the current generated
a demand-side management approach was proposed through
at node n, phase f , time t
im re a day-ahead optimization process for ESDs and distributed
Imn,f,t , Imn,f,t Imaginary and real part of the current in
generation (DG). This formulation only considered end users,
circuit mn, phase f , time t
SDim SDre disregarding the economic and operational constraints.
Iu,t , Iu,t Imaginary and real part of the current de-
Lately, the number of electric vehicles (EVs) connected
manded by ESD u, time t
D to EDNs has received more attention. Because the EVs use
Pn,f,t , QD n,f,t Active and reactive power demand at node n,
the energy from the grid to charge their batteries, a high
phasef , time t
G penetration of EVs can cause the EDN to work outside of
Pn,t , QG n,t Active and reactive power generated at node n,
acceptable operational limits [12]. The development of EV
time t
EV charging schedules to prevent issues such as overloads and
Pe,t EV e active power consumption, time t
SD+ SD− voltage limit violations is essential [13]. Several works have
Pu,t , Pu,t Charging and discharging power of the ESD u,
approached the EV charging coordination (EVCC) problem
time t
[13]– [16]. In [13], an EVCC problem approach was presented.
Qcbn,t Reactive power delivered by a CB at node n, time t
The formulation could properly handle large populations and
tpmn,f,t Integer variable for the number of steps of the VR
the random arrival of EVs, but the impact of the EVs on
in circuit mn, phase f , time t
im re the grid was not considered. A mathematical model for EV
Vn,f,t , Vn,f,t Imaginary and real part of the voltage at node
integration into the grid was presented in [14]. An economical
n, phase f , time t
c(re) evaluation was assessed in order to find the optimal energy cost
Vmn,f,t,k Real part of the VR voltage calculation auxiliary that would benefit both EV owners and the EDN operator. In
variable in circuit mn, phase f , time t, position k [15], a multi-period optimization for EV charging in EDNs
ye,t Binary variable associated with the charging state of was presented. The optimization aimed at minimizing the cost
EV e, time t of charging EVs. An algorithm based on sensitivities was
proposed for the real-time EVCC in [16], considering the
I. I NTRODUCTION random arrivals/departures of EVs, voltage profiles, and power

I NCREASING energy efficiency and reducing energy costs

through a Volt VAr control (VVC) in electrical distribution
networks (EDNs) has received plenty of attention in recent
years [1]– [4]. The VVC problem finds a control scheme for
the proper operation of the devices in the EDN that manage
variations in voltage magnitudes and reactive power flows.
These devices include among others, voltage regulators (VRs),
capacitor banks (CBs) and on-load tap changers (OLTCs)
[5]– [8]. Recently, several VVC methodologies have been
presented in the specialized literature. In [1], it is proposed
a centralized function for VVC, in which, a state estimation
based on pseudo-measurements using neural networks is used
generation limits in order to minimize the total energy cost.
The presented approach overcomes the shortcomings of
previous studies, which solved the VVC and the EVCC
separately, by engaging both problems in a more global control
optimization of the EDN. Furthermore, the advantages offered
by ESDs and dispatchable DG units are also considered in
this formulation. While VVC devices and ESDs improve
the operation of the EDN, the formulation seeks to avoid
violations of operational limits and reduce the overall costs
in the grid. In addition, the EVCC tries to optimally charge
the EVs plugged into the EDN, while avoiding the issues that
a high EV penetration may cause.

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy
This work proposes an approach that simultaneously uses
the VVC (management of OLTCs, VRs, and CBs), ESD
operation, and dispatchable DG units to improve the EVCC
in a three-phase unbalanced EDN. Different from the single
phase equivalent, the three-phase unbalance representation
allows the inclusion of the mutual coupling effects in the
circuits, reaching a more realistic representation. Moreover,
in order to develop a thorough analysis and a more accurate
economic improvement, the load’s voltage dependency is also
taken into account.
In order to integrate the described devices and achieve
the aforementioned objectives, a mixed-integer nonlinear pro-
gramming (MINLP) formulation is found. Linearization tech-
niques are later used to obtain a mixed-integer linear pro-
gramming (MILP) model that can be solved using commercial
software, guaranteeing the optimal solution to the problem.
A step-by-step methodology that offers a broad view of the
EDN’s behavior in future periods is used to solve a 24h
time period. The proposed formulation was tested in an EDN
with 34 medium-voltage nodes and 144 low-voltage nodes.
The results demonstrate the efficiency and robustness of the
methodology. This work’s main contributions are as follows:
• A step-by-step methodology that solves the EVCC prob-
lem and finds the optimal operation of the EDN, through-
out a centralized coordination which takes into account
the benefits and flexibility brought to the grid by VVC
devices, ESDs, and dispatchable DG units.
• A multi–period MILP formulation to solve the EVCC
problem together with VVC, ESD operation and dispatch-
able DG units in a three-phase unbalanced EDN, in which
loads are represented via a voltage dependent load model.
II. P ROBLEM F ORMULATION AND S OLUTION T ECHNIQUE
This section describes the EVCC problem as well as the
advantages and all technical concerns related to the VVC and
the ESD operation. A step-by-step methodology proposed to
solve the EVCC, taking into account VVC, ESD operation,
and dispatchable DG units in a three-phase unbalanced EDS
and using a voltage dependent load model, will be presented.
A. EVCC, VVC and ESD scheduling
1) EVCC: The EVCC problem aims to find an optimal
schedule for the battery charging of the EVs connected into
the EDN in a specific time period. This charging scheduling
not only has to satisfy the EV owner needs but also must avoid
technical limit violations due to excessive EV load connected
into the grid. An improvement in the EVCC can lead to:
An economic betterment in the operation of the system, i.e.,
reduction in energy losses; an amelioration in the fulfillment
of technical limits; a cost reduction in the EV recharge; or a
reduction in the energy curtailment of the EVs at departure.
For a solution to the EVCC problem, the following consid-
erations are assumed:
• The initial SOC of every vehicle is known when the
vehicle is plugged into the grid.
• The batteries can be controlled in each time interval into
which the time period is divided.
1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
• The EV user may provide a departure time when the
vehicle is plugged; otherwise, it will be assumed that
the vehicle will remain plugged in until a probable time
interval, determined by the historical of departure of the
EV.
2) VVC: The VVC is defined as a control strategy to
manage the voltage magnitudes and the reactive power flow
throughout the EDN. This management consists of periodic
adjustments over devices that inject reactive power into the
grid to improve the voltage drop and devices that directly
control voltage magnitude [7], [8]. In general, the VVC defines
control actions in OLTCs, VRs, and fixed–switchable CBs.
To properly carry out the adjustment over VVC devices,
the control actions might be handled by a managed operation.
This operation often follows local decision–making based
on local measurements of the EDN, i.e., each device is
controlled independently. However, each action affects the
EDN operation and the performance of other devices. Thus,
an integrated coordination of all VVC devices, based on
remote measurements on the EDN and operator knowledge, is
necessary to obtain an optimal EDN operation through VVC.
For centralized coordination, a communication infrastructure
and an optimized decision-making algorithm is required to
achieve an optimal integrated VVC [17].
Keeping voltage magnitudes inside acceptable limits while
optimizing energy loss reduction has been the classical ap-
proach when solving the VVC problem. Nonetheless, in recent
works the VVC is also used to adjust the voltage profile to
manage total demand since a reduction in voltage magnitude
levels can represent a decrease in power consumption [2].
Therefore, considering the effects of special loads in the EDN
operation, the VVC can be a useful tool to improve the solution
of the new challenges emerging in the EDN operation. Hence,
an optimized dynamic coordination of the grid can be obtained
with the inclusion of the VVC, allowing a more flexible
adjustment of the EDN, while avoiding voltage limits breaches
in the presence of special loads such as EVs.
3) ESD operation: As mentioned, the solution of the pro-
posed methodology must define an optimal charging agenda
for the ESDs. For the ESDs, the charging schedule is responsi-
ble for throttling the battery charge and for allowing the EDN
to use the stored energy. Thus, the formulation must determine
the optimal schedule to charge and discharge the ESDs at each
time interval. It is assumed that:
• The ESDs cannot be discharged beyond their specified
Depth of Discharge (DoD).
• The ESDs can be controlled through communication
devices in order to define the charging state at each time
step.
• For each ESD, a goal is defined for the SOC at the end
of the time period (e.g., same as the initial SOC). This
goal is used to represent the utilization of the ESDs after
the considered time period (e.g., one day ahead).
B. Voltage Dependent Load Model
The demand of an EDN is classified into loads that can be
represented as: a) constant power loads, b) constant impedance
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy
INPUTS
Relax Integrality
START for all Decision t* = 1
Variables
EV Historical
Profiles
Updated Data
for Connected
EVs
t* = t* + 1
Figure 1. Step–by–step flowchart of the proposed methodology.
loads, c) constant current loads, or d) a combination of these.
In a VVC framework for EDNs, the load’s voltage dependency
is a key aspect of the suitable representation of the network
operation. As discussed in [2], [3], and [18], VVC models are
highly dependent on the accuracy of the load representation.
Dependence on the voltage magnitude and frequency is consid-
ered in the load models; mathematically, this dependence can
be represented by static and dynamic load models described by
the traditional ZIP model. For this analysis, two static models
are commonly studied: the polynomial load model and the
exponential load model [19]. The exponential expression of
the load’s sensitivity to the voltage and frequency variation is
represented by (1) and (2).
 α
V estimated time interval, a different estimation may be used or
PD = Po (1 + Kpf (fr − fro )) (1)
Vo
 β
D o V the EVs may arrive and depart during any time interval of
Q =Q (1 + Kqf (fr − fro )) (2)
Vo
where, fr represents the frequency of the bus voltage and
fro represents the nominal frequency. The coefficients Kpf
and Kqf are the frequency sensitivities for the active and
reactive power loads, respectively. In this work, the effects
associated with frequency are disregarded. Furthermore, with
an appropriate adjustment of the constants α and β the
model can be restricted to the steady-state analysis case (i.e.,
dependence directly on the voltage magnitude). Appropriate
values for these constants may be found in previous works,
such as [20].
C. Methodology
The proposed methodology defines the status of the control
devices (VRs, and CBs) as well as the charging status of the
EVs in the network by solving an optimization problem. The
operation of the EDS is modeled considering a given horizon
(e.g. one-day), which is divided in time intervals. The time
duration for every time interval is represented by ∆t . The set
T orders time intervals into which the time period is divided.
1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
STEP-BY-STEP
OPTIMIZATION
OUTPUTS
Update EV Parameters
(SOC, Arrival, Departure)
Control Actions for:
- Connected EVs
Fix Decision Variables for - VVC Devices
all Time Intervals before t* - ESDs
- DG Units
Enable Integrality of the
Decision Variables in t*
No
t* < 24 END
Solve Optimization
Problem Yes
As shown in Fig. 1, the proposed MILP model is solved at
the beginning of each time interval considering the actual num-
ber of connected EVs; besides, a forecast for the parameters
related to possible EV plugs, such as estimated arrival times
and initial state of charge (SOCs), is also taken into account.
A dynamic scheduling (rolling multi-period optimization for
the EVCC) is performed determining at each time interval
the optimal charging agenda of the EVs plugged and to be
plugged into the grid. This dynamic scheduling is a part of the
step-by-step methodology proposed, which will also determine
at each time interval the control scheme for VVC devices,
DG units and ESDs on the remaining time period. When
a forecasted EV is actually plugged in, the parameters are
updated. Nevertheless, if an EV does not arrive during the
the EV may be disregarded altogether. It should be noted, that
the time period. Therefore, the EV’s charging agenda will be
generated between arrival and departure, ideally dispatching a
fully charged battery for every vehicle. Different techniques
may be applied to estimate the arrivals, departures, and initial
SOCs of the EVs, as presented in [22]– [24].
Operational constraints such as voltage limits, active and
reactive power generation limits, and maximum current limits,
are taken into account in the model. This way, a step-by-
step solution is constructed over the entire time period. The
solution at each time interval will determine the charging
state for the EVs and ESDs, the energy delivered by the DG
units, the number of modules operating in each CB, and the
tap position for each phase of every VR. The nature of the
control variables associated with each device is binary, which
makes the VCC problem hard to solve. In order to reduce the
computational complexity, only the control variables for the
current time interval are treated as binary variables, i.e., control
variables for the remaining time intervals are relaxed. This
strategy makes it possible to solve the EVCC problem with
reduced computational effort, obtaining the solution for the
current time interval using updated information. An important
upside of this methodology is that it offers a broad view of
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

q
the state of the EDN throughout the whole time period. QD
n,f,t = Q o
n,f,t ( re2 + V im2 /V )βn,f
Vn,f,t n,f,t o (9)
X
re re re im
Vm,f,t − Vn,f,t = (Rmn,f,h Imn,h,t − Xmn,f,h Imn,h,t )
h∈F
(10)
D. Mixed-Integer Nonlinear Programming Model for EVCC im im
X
re im
Vm,f,t − Vn,f,t = (Xmn,f,h Imn,h,t + Rmn,f,h Imn,h,t )
h∈F
A mixed integer nonlinear programming (MINLP) model (11)
is presented in (3)–(38) for the proposed EVCC problem. 2
2 re 2 im 2
The solution found by the formulation must define: the power V ≤ Vn,f,t + Vn,f,t ≤V (12)
G 2 2
delivered by each dispatchable DG unit (Pn,t ), the number of re
0 ≤ Imn,f,t 2 im
+ Imn,f,t ≤ I mn (13)
modules connected in each CB (Bn,t ), the tap position for each
VR (tpmn,f,t ), the charging schedule for each EV (yn,t ), and ∀mn ∈ L, ∀n ∈ N, f ∈ F, t ∈ T
the charging and discharging schedule for each ESD (PuSD+
where, F , L, and N , are the sets of phases, circuits and
and PuSD− , respectively).
nodes, respectively. The balance of the currents at each node
1) Objective Function (OF): Represented by (3), the OF is represented by (4) and (5). Moreover, γ is a binary function
aims to minimize the cost of the energy provided by the that takes a value of 1 if the EV e or ESD u are connected at
substation (SE) (first term), the DG energy cost (second term), node m and at phase f . The load currents shown in (6) and
the cost of the energy curtailment on EVs (third term), and the (7) define the relationship between voltage, current and the
cost of the energy injected from the ESDs to the grid (fourth active and reactive powers demanded by the loads. Moreover,
term), and to penalize the deviation from the defined goals for (8) and (9) are the exponential representation of the load’s
the SOC of the ESDs (fifth term). voltage dependency. Equations (10) and (11) are the result of
applying Kirchhoff’s Voltage Law to each independent loop
XX
G re Gre im Gim

of the EDN. Finally, the limits for the voltage magnitude and
min ζSE,t ∆t VSE,f,t ISE,f,t + VSE,f,t ISE,f,t +
current capacity in each circuit are stated in (12) and (13),
f ∈F t∈T
XX X respectively.
G G
ζn,t ∆t Pn,t + ζcEV EeEV c + 3) DG Constraints: The operation limits of the DG units
n∈N t∈T e∈Ξ are defined by (14)–(16), while the active and reactive power
X X X
SD SD−
ζ ∆t Pn,t + ζcSD EuSDc (3) injections are represented by (17)–(18), respectively.
u∈SD t∈T u∈SD
G G
0 ≤ Pn,t ≤ Pn (14)
G
2) Load Flow Constraints: The equations (4)–(13) model QGn
≤ QG n,t ≤ Qn (15)
the steady-state operation of a three-phase unbalanced EDN |QG G
n,t | ≤ Pn,t tan(arccos(φn )) (16)
considering the presence of DG units (set DG), CBs (set CB), G re Gre im Gim
Pn,t /3 = Vn,f,t In,f,t + Vn,f,t In,f,t (17)
VRs (set V R), ESDs (set SD) and EVs (set Ξ).
QG re Gim im Gre
n,t /3 = −Vn,f,t In,f,t + Vn,f,t In,f,t , (18)
Gre
X
cbre
X
re
X
re ∀n ∈ N |n∃DG , f ∈ F, t ∈ T
In,f,t + In,d + Ikm,f,t − Imn,t −
n∈CB km∈L∪V R mn∈L
im
4) CBs Constraints: Similar to the DG units, the active and
X
sh
X
sh
Vm,f,t Dre reactive power of the switchable CBs are represented by (19)
( Bkm,f + Bmn,f ) = Im,f,t +
2 and (20). The value for the active power injection of every
km∈L mn∈L
X
EV re
X
SDre CB will always be equal to zero. Equation (21) represents the
Ie,t γ(e, m, f ) + Iu,t γ(u, m, f ) (4)
number of capacitor modules connected to the grid for each
e∈Ξ u∈SD
X X X bank, while the allowable limit of modules is modeled by
Gim cbim im im
In,f,t + In,d + Ikm,f,t − Imn,t − (22). Finally, as the lifetime of switchable CBs is significantly
n∈CB km∈L∪V R mn∈L impacted by the number of operations (23), this limits the
re
X
sh
X
sh
Vm,f,t Dim maximum number of operations permitted over the entire time
( Bkm,f + Bmn,f ) = Im,f,t +
2 period.
km∈L mn∈L
X X
EV im SDim
Ie,t γ(e, m, f ) + Iu,t γ(u, m, f ) (5) re
0 = Vn,f,t cbre
In,f,t im cbim
+ Vn,f,t In,f,t (19)
e∈Ξ u∈SD
2 2 Qcb
n,t /3
re
= −Vn,f,t cbim
In,f,t + im cbre
Vn,f,t In,f,t (20)
Dre D re
In,f,t = (Pn,f,t Vn,f,t + QD im re im
n,f,t Vn,f,t )/V n,f,t + V n,f,t (6) cb
2 2
Qn,t = Bn,t Qesp
n (21)
Dim D im
In,f,t = (Pn,f,t Vn,f,t − QD re re im
n,f,t Vn,f,t )/V n,f,t + V n,f,t (7) 0 ≤ Bn,t ≤ B n (22)
q
D o re2 + V im2 /V )αn,f
Pn,f,t = Pn,f,t ( Vn,f,t
X
n,f,t o (8) |Bn,t − Bn,t−1 | ≤ ∆cb (23)
t∈T
∀n ∈ N |n∃CB , f ∈ F, t ∈ T

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

5) VR and OLTC Constraints: The mathematical model of E. Mixed-Integer Linear Programming Model for EVCC
a VR or a OLTC is shown in (24)–(27). In this model, (24)
represents the real regulated voltage and (25) represents the To enhance the robustness of the solution and to enable
real regulated current on each VR/OLTC. Expression (26) rep- the utilization of classic optimization methods and commercial
resents the minimum and maximum limits of the tap position solvers, linearization techniques are applied to the nonlinear
represented by the integer variable tpmn,f,d . Analogue to (23), expressions in (3)–(38) in order to obtain a MILP formulation.
(27) limits the maximum number of operations allowable for 1) Linearization for Power Flow Nonlinear Constraints:
a VR/OLTC over the time period in order to avoid excessive Constraints (6)–(9), (12), and (13), are nonlinear expressions.
wear and tear of these devices. Expressions similar to (24) and Equations (6)–(9) are rewritten as shown in (39) and (40). Con-
(25) were implemented to describe the imaginary component straints (41) and (42) show the first-order linearization for (39)
re∗ im∗
for both the regulated voltage and current. and (40) around an estimated operation point (Vn,f,t , Vn,f,t ).
Furthermore, (12) and (13) are linearized as presented in [21].
re re
Vn,f,t = (1 + %Rmn tpmn,f,t /T pmn )Vm,f,t (24)
re re
re
Vn,f,t  2 2
 αn,f
2 −1
Ikm,f,t = (1 + %Rmn tpmn,f,t /T pmn)Imn,f,t (25) re
g(Vn,f,t im
, Vn,f,t o
) = Pn,f,t re
Vn,f,t im
+ Vn,f,t
Vo αn,f
− T pmn ≤ tpmn,f,t ≤ T pmn (26) im   βn,f
Vn,f,t re2 im2 2 −1
Qon,f,t
X
|tpmn,f,t − tpmn,f,t−1 | ≤ ∆vr (27) + Vn,f,t + Vn,f,t (39)
t∈T
Vo βn,f
im  αn,f
Vn,f,t  −1
∀mn ∈ L|mn∃V R , f ∈ F, t ∈ T re
h(Vn,f,t im
, Vn,f,t o
) = Pn,f,t re2 im2 2
αn,f Vn,f,t + Vn,f,t
Vo
6) ESD Constraints: For the ESDs, (28) and (29) represent re 
Vn,f,t  βn,f
2 −1
o re2 im2
the active and reactive power exchange between each ESD and − Qn,f,t βn,f Vn,f,t + Vn,f,t (40)
Vo
the grid. In (30), the energy balance in an ESD is written in ∗ ∗
∂g  re∗ ∂g  im∗
terms of the energy that must be charged, the charging power Dre
In,f,t = g∗ + re
re
(Vn,f,t − Vn,f,t )+ im
im
(Vn,f,t − Vn,f,t )
in each time interval and the defined goal for the SOC at the ∂V  ∂V 
end of the charging period. Equation (31) defines the energy (41)
∗ ∗
stored in each SD at every interval and (32) limits the ESD’s ∂h  re∗ ∂h  im∗
Dim
In,f,t = h∗ + re
(Vn,f,t − Vn,f,t )+ im
(Vn,f,t − Vn,f,t )
energy level ensuring that it does not exceed the battery’s ∂V re  ∂V im 
maximum capacity or the pre-established DoD. Finally, (33) (42)
defines the limits for the charging and discharging active ∀n ∈ N, f ∈ F, t ∈ T
powers for each ESD.
SD+ SD− re SDre im SDim 2) Linearization of DG, CBs, ESDs and EVs Nonlinear
(Pu,t − Pu,t )/3 = Vu,t Iu,t + Vu,t Iu,t (28)
re SDim im SDre
Constraints: The nonlinear expressions of the DG, CBs, ESDs
0 = −Vu,t Iu,t + Vu,t Iu,t (29) and EVs (constraints (17), (18), (19), (20), (28), (29), (34)
SD SD
Eu f = Eu,T + EuSDc (30) and (35)) are approximated using an estimated operation point
re∗ im∗
SD SDi
X
SD+ + SD− − (Vn,f,t , Vn,f,t ) as shown in [21]. The quality of the estimated
Eu,t = Eu + ∆k (Pu,t ηu − Pu,t /ηu ) (31) operation point will define the approximation error. Historical
k∈T,k≤t
data and the EDN operator knowledge, nominal values for the
SDi SD SD SD
min(Eu , E u DoD) ≤ Eu,t ≤ Eu (32) voltage magnitudes, or a linear relaxation of the MILP model
SD+ SD− SD initially solved (in which the binary nature of the decision
0 ≤ Pu,t , Pu,t ≤ Pu ∀u ∈ SD, t ∈ T (33)
variables is temporarily ignored), could be used to estimate
7) EV Constraints: The active and reactive powers of the operation point.
the EVs are presented in (34) and (35), respectively. The 3) Linearization of VR/OLTC Nonlinear Constraints: In
instant power in (36) is represented in terms of the maximum the expressions (24) and (25), the product of the decision
charging power and the binary control variable yn,t . Constraint variables tpmn,f,t and, Vm,f,t or Imn,f,t on the real and
(37) determines the energy curtailment on each EV, as (38) imaginary components is linearized. To obtain linear expres-
determines the energy of each EV at departure. The variable sions, the integer number of steps is represented as a set
yn,t has a value of 1 if the EV battery is charging at its of binary variables btmn,f,t and the products tpmn,f,t Vm,f,t ,
c
maximum power P e
EV ch
and a value of 0 otherwise. tpmn,f,t Imn,f,t are substituted by the variables Vmn,f,t,k and
c
Imn,f,t,k , respectively, on the real and imaginary components.
EV re EV re im EV im
Pe,t = Ve,t Ie,t + Ve,t Ie,t (34)
re EV im im EV re 2T pmn
0 = −Ve,t Ie,t + Ve,t Ie,t (35) X %Rmn c(re)
re re
Vn,f,t = (1 − %Rmn )Vm,f,t + V (43)
EV
Pe,t = P e
EV ch
ye,t (36) T pmn mn,f,t,k
k=1
EV 2T pmn
E e = Ee + EeEV c
EV
(37) re re
X %Rmn c(re)
Ikm,f,t = (1 − %Rmn )Imn,f,t + I
EeEV = Eeini +
X
EV
∆t Pe,t ηe ∀e ∈ Ξ, t ∈ T (38) T pmn mn,f,t,k
k=1
t∈T (44)

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

2T pmn xx11 xx09

X S

btmn,f,t,k − T pmn = tpmn,f,t (45) 01

xx xx01 xx02 xx03 xx05 xx06 xx07 xx08
k=1 02
c(re) xx12 xx10
re
|Vm,f,t − Vmn,f,t,k | ≤ V (1 − btmn,f,t,k ) (46) 04
03
14

c(re) 31 33
|Vmn,f,t,k | ≤ V btmn,f,t,k (47) 05 11 22 28

06 09 20 26 CB
re c(re)
|Imn,f,t − Imn,f,t,k | ≤ I mn (1 − btmn,f,t,k ) (48) VR

07 08 10 12 15 16 17 19 21 23 25 27 29 32
c(re)
|Imn,f,t,k | ≤ I mn btmn,f,t,k (49) DG + -
13 18 24 ESD 30

btmn,f,t,k ≤ btmn,f,t,k−1 (50)

∀mn ∈ L|mn∃V R , f ∈ F, t ∈ T, k ∈ 1..2T pmn Figure 2. 178-nodes test system.
Table I
Expressions (43)–(50) are linear extensions of (24)–(27), H OURLY E NERGY C OSTS AND L OAD VARIATIONS
where (43) represents the calculation of the regulated voltage
and (44) represents the calculation of the regulated current Hour Energy cost % of peak Hour Energy cost % of peak
($/kWh) load ($/kWh) load
for the real component. Equation (45) associates the binary 08:00 0.0442 50 20:00 0.0875 89
variable with the tap integer variable. Constraint (46) defines 09:00 0.0529 55 21:00 0.0837 87
c 10:00 0.0606 63 22:00 0.0721 75
the auxiliary variable Vmn,f,d,k and (47) describes its limits. 11:00 0.0673 70 23:00 0.0625 69
c
In the same way, (48) defines the auxiliary variable Imn,f,d,k 12:00 0.0721 75 00:00 0.0529 59
and (49) describes its limits. Expression (50) represents the 13:00 0.0798 83 01:00 0.0385 46
14:00 0.0847 88 02:00 0.0327 44
sequencing of the binary variable btmn,f,d in the previous tap 15:00 0.0866 90 03:00 0.0317 42
position. For time intervals in which the binary nature of the 16:00 0.0914 95 04:00 0.0336 41
17:00 0.0933 97 05:00 0.0327 40
control variables is disregarded, the equations above are not 18:00 0.0962 100 06:00 0.0394 45
suitable. Therefore, (51) and (52) are used to calculate the 19:00 0.0895 91 07:00 0.0423 48
relaxed value of bt for each VR/OLTC.
In other words, (43)–(50) are used to calculate the tap interval during the period in bold, based on the fact that this
position in the actual time interval and, (50), (51) and (52) are is the time window when the EV demand has a greater impact.
constraints applied to the rest of the time period. Similar to the G
The DG unit energy cost ζn,t was equal to 0.045 US$/kWh,
approximated values of the voltages, the quality of the approx- while its maximum active power was 350 kW and its minimum
re∗ im∗
imated values of the currents (Imn,f,t and Imn,f,t ) depends on and maximum reactive powers were equal to -175 kVAr and
the EDN operator’s knowledge. Expressions similar to (43)– 175 kVAr, respectively. Finally, the minimum power factor for
(52) were implemented to linearize the imaginary component. the operation was 0.90. The ESD was assumed to be owned by
2T pmn the EDN operator. Hence, the cost for the active power injected
re re
X %Rmn re∗ into the grid by the ESD ζ SD was disregarded. Moreover, the
Vn,f,t = (1 − %Rmn )Vm,f,t + btmn,f,t,k Vn,f,t
T pmn ESD maximum energy capacity was 3 MWh, with a maximum
k=1
(51) charge and discharge power of 300 kW, 10% of DoD , and
2T pmn charging and discharging efficiencies of 100%. The expected
re re
X %Rmn re∗ ESD SOC at the end of the time period was set at its initial
Ikm,f,t = (1 − %Rmn )Imn,f,t + btmn,f,t,k I
T pmn mn,f,t energy level. Finally, the curtailment costs ζcSD was set to
k=1
(52) 100 US$/kWh.
∀mn ∈ L|mn∃V R , f ∈ F, t ∈ T The VR was assumed to be a three-phase VR built from
three single-phase VRs. This feature offers the flexibility of
controlling each phase independently. Each single-phase VR
III. T ESTS AND R ESULTS could vary the voltage with a regulation ratio of ±10%;
To assess the proposed methodology, tests were carried in addition, this ratio was equally distributed across 16 tap
out on a three-phase EDN with 34 nodes at medium voltage positions (±8). Furthermore, the switchable CB capacity was
(24.9 kV) and 144 nodes at low voltage (480 V) shown in 300 kVAr, divided into 6 individual modules. For these tests,
Fig. 2. The EDN was adapted from the IEEE 34-node test the maximum number of allowable operations for both CB
system; for further information, the complete description of and VR (∆cb and ∆vr , respectively) were set at unreachable
the EDN used is available in [25]. The test system includes values, i.e., the number of operations for each device over the
a DG unit at node 8, an ESD at node 23, a VR connected at time period was unlimited.
the end-point of circuit 16-17 and a CB at node 31. Moreover, In addition, 180 consumers with EVs were considered in the
phases A, B, and C were loaded with 33%, 35%, and 32% of network, representing almost a 40% penetration. Two types of
the total demand, respectively. The values of the parameters α EVs were considered: the ‘Tesla Model S’ with a 70 kWh
and β, used to represent different types of loads in the voltage battery capacity and 10 kW charging power [26], and the
dependent model, are presented in [25]. The hourly energy ‘Nissan Leaf’ with a 24 kWh battery capacity and 4 kW
costs and load variation percentages are shown in Table I. It charging power [27]. Moreover, a 100% efficiency for all EV
was assumed that EVs could arrive and depart at any time charging stations was assumed and the curtailment cost for the

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

Table II (a)
S UMMARY O F T HE T EST C ASES 2000

Active Power (kW)

SE DG Total EV Energy Objective
Case Energy Energy Energy Curtailment Function 1000
Cost ($) Cost ($) Cost ($) (kWh) ($)
I 1957.69 314.98 2272.67 274 29672.67 0
II 1969.79 336.18 2305.97 104 12705.97
III 1905.24 279.13 2184.37 0 2184.37
IV 1852.81 287.94 2140.75 0 2140.75 -1000

08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 00 01 02 03 04 05 06 07 08
EVs ζcEV was set to 100 US$/kWh. Hour
The proposed MILP model was written in the mathematical
(b)
language AMPL [28] and solved using the commercial solver 2000

Active Power (kW)

CPLEX [29], on a computer with an Intel i7 4770 proces-
sor. Tests were carried out for four different cases with the 1000

following control alternatives:

0
• Case I: ESD not available; VVC predefined.
• Case II: ESD enabled; VVC predefined. -1000
• Case III: ESD not available; VVC enabled.
08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 00 01 02 03 04 05 06 07 08
• Case IV: ESD enabled; VVC enabled.
Hour
For the Cases I and II, in which the VVC was predefined, (c)
the VVC devices followed preset settings. These settings are 2000

Active Power (kW)

defined by the EDN operator following the daily conventional
1000
demand pattern.
Table II shows a summary of the results from the test cases. 0
Case I, which includes only the control over the EVs plugged
into the system, presented the worst solution, with high EV -1000
energy curtailment and expensive overall energy cost. From
08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 00 01 02 03 04 05 06 07 08
Case II, it can be seen how the enabling of the ESD in the grid Hour
leads to a significant reduction in the EV energy curtailment,
(d)
hence, an improvement in the EVCC solution. Besides, Case 2000
Active Power (kW)

II presented an increment on the overall energy costs due to

1000
a higher EV demand attended. Moreover, Case III shows how
with the inclusion of the VVC, the grid is able to attend the 0
whole EV demand, avoiding technical limit breaches. Finally,
-1000
the best OF value with the lowest overall energy cost and no
EV energy curtailment was obtained in Case IV, demonstrating 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 00 01 02 03 04 05 06 07 08
the advantages of the proposed control methodology. Power Power Power
Hour
Power Power
The time limit for the solution process at each time interval delivered delivered delivered demanded demanded
by the SE by the DG by the ESD by the EVs by the ESD
was 600s. If the time limit was reached and the optimal
solution had not been found, the best integer solution yet Figure 3. Active power injected into the grid for a) Case I, b) Case II, c)
found was adopted. Furthermore, it was verified that the gap Case III and d) Case IV
presented in those time intervals, in which the solver could
not find the optimal solution within the specified time limit, as well as the energy bought from the DGs.
never exceeded 0.1%. The VVC pre-set profile and the VVC schemes for Cases
Fig. 3 shows the active power injected into the grid by the III and IV are shown in Table III. For every case, the VVC
SE, the DG unit, and the ESD for Cases I–IV, respectively. In seeks to fulfill the technical constraints and determine the
addition, this figure shows, for all cases, the power demanded best economical operation of the EDN. When enabled, all the
by the EVs and the ESD when charging (by convention, the modules of the CB were plugged into the grid; this permits
active power due to these demands is shown in negative val- a greater voltage reduction at the VR, therefore, achieving
ues). It should be noted that the methodology found an optimal a better economic performance. Although the limit for the
schedule to recharge the EVs mainly at low-cost energy time number of switching operations was set at a very large number,
intervals. Nevertheless, energy curtailments were presented the number of switching operations for any VVC device over
when the grid was unable to meet the EV demand. When the entire time period for Cases III and IV was below 30.
VVC was enabled, the EDN could completely charge the EVs Fig. 4 shows the minimum voltage magnitude profiles for
during the low-cost energy time intervals, thereby achieving a each time interval in the EDN, evidencing the fulfillment of
better OF. On the other hand, the ESD improves the EVCC voltage limits in all cases. Due to the voltage dependency of
by allowing a higher number of EVs to be scheduled at the the loads, in the cases where the VVC was enabled (Cases
last time intervals. Furthermore, as the VVC was enabled, the III and IV), the minimum voltage profile was kept as close as
total amount of power demanded in the time period decreased possible to the lower boundary. This feature can be quantified

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

Table III -4
18:00
-2
01:00
10 Phase A 10 Phase A
VR TAP P OSITION R ANGE FOR EACH P HASE AND CB M ODULES 5 6

Time intervals 07–12 13–21 22–01 02–06

Error Percentage (%)

0 0
20 40 60 80 100 120 140 160 178 20 40 60 80 100 120 140 160 178
Predefined A,B,C {2} {4} {2} {1} -4
10
-2
10
Phase B Phase B
Settings CB {3} {5} {3} {1} 5 6

A {-6,-3} {-3,-1} {-4,4} {-1,4}

Case III B {-4,-2} {-1,1} {2,6} {-5,5}
C {-6,-4} {-3,-2} {-1,0} {-1,5} 0
20 40 60 80 100 120 140 160 178
0
20 40 60 80 100 120 140 160 178
CB {6} {6} {6} {6} -3
10 Phase C
-2
10 Phase C
6 6
A {-5,-3} {-3,-1} {-4,4} {-2,3}
Case IV B {-4,-2} {-1,0} {1,5} {-3,5}
C {-5,-4} {-3,-2} {-1,2} {-6,4}
0 0
CB {6} {6} {6} {6} 20 40 60 80 100 120 140 160 178 20 40 60 80 100 120 140 160 178
Nodes

0.98 Figure 5. Voltage magnitude error percentage between the MILP model
Minimum Voltage (pu)

0.96
0.94
0.92
0.90
08 10 12 14 16
Figure 4. Minimum voltage profile
by the economical operation improvement shown in Table II.
Thereby, comparing Cases I and III, there is a reduction in
the overall energy cost of the EDN throughout the entire time
period, even though the EVs were completely attended only
in Case III.
To properly validate the importance of the voltage de-
pendent load model, the formulation was solved considering
constant power in the load representation (i.e., α = 0 and
β = 0). A power flow considering the voltage dependent
load model was later executed, fixing the control variables
obtained from the constant power test. The results showed
that the solutions obtained using an approach without proper
load modeling breached technical limits when applied to the
grid. Furthermore, the solution obtained from the constant
power test not only exceeded operational constraints, but also
presented higher OF.
Finally, in order to evaluate the precision of the proposed
formulation, all the decision variables of the solution found
by the methodology [i.e., the power delivered by each dis-
G “Neural networks to improve distribution state estimation Volt-VAr
patchable DG unit (Pn,t ); the number of modules connected
in each CB (Bn,t ); the tap position for each VR (tpmn,f,t );
the charging schedule for each EV (yn,t ); and the charging
and discharging schedule for each ESD (PuSD+ and PuSD− ,
respectively)] in Case IV were used to solve a conventional AC
power flow. Thereby, for the OF, the error found by comparing
the MILP formulation and the conventional power flow was
0.023%. Moreover, the voltage magnitude error percentage
obtained for all nodes at the time intervals with the highest
conventional load and EV demand (hours 18:00 and 01:00,
respectively), is presented in Fig. 5. It is worth noting that
this error percentage does not exceed 0.005% at 18:00 hours
and 0.06% at 01:00 hours. Hence, these results validate the
presented MILP formulation as an accurate approximation of
the EDN steady-state operation.
Case I solution in Case IV and the conventional power flow for time intervals 18:00
Case II
and 01:00.
Case III
Case IV IV. C ONCLUSION
A new methodology based on a mixed-integer linear pro-
gramming model for the electric vehicle charging coordination
(EVCC) problem considering Volt VAr control (VVC), energy
storage device (ESD) operation, and dispatchable distributed
18 20 22 00 02 04 06
Hour generation (DG) units in a three-phase unbalanced electrical
distribution network (EDN), taking into account a voltage de-
pendent load model, was presented. The proposed formulation
was tested on an EDN with 34 medium–voltage nodes and
144 low–voltage nodes. The results showed that the optimized
operation of the DG, ESD, VVC devices, and EV recharge
represented an overall energy cost reduction and guaranteed
the avoidance of technical limit violations.
The dynamic coordination was proved to be efficient, be-
cause it defines the step to be implemented and gives a broad
view of the EDN state throughout the whole time period.
Moreover, this methodology handles the randomness in the
electric vehicle (EV) arrival and departure times, initial state
of charge, battery sizes, and forecast errors.
The presented results highlight the importance of the pro-
posed methodology, which encompasses the optimal control
of the aforementioned devices. Besides, they foreground how
the ESDs help in the avoidance of energy curtailment in EV
recharging and how the VVC scheme becomes more suitable
in order to avoid voltage limit violations in the EDN.
R EFERENCES
[1] M. Biserica, Y. Besanger, R. Caire, O. Chilard, and P. Deschamps,
control performances,” IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 1137-
1144, Sep. 2012.
[2] A. Padilha-Feltrin, D. Quijano, and J. R. Mantovani, “Volt-VAr mul-
tiobjective optimization to peak-load relief and energy efficiency in
distribution networks,” IEEE Trans. Power Del., vol. 30, no. 2, pp.
618-626, Apr. 2015.
[3] H. Ahmadi, J. R. Marti, and H.W. Dommel, “A framework for Volt-
VAr optimization in distribution systems,” IEEE Trans. Smart Grid,
vol. 6, no. 3, pp. 1473-1483, May 2015.
[4] K.P. Schneider and T.F. Weaver, “A method for evaluating Volt-VAr
optimization field demonstrations,” IEEE Trans. Smart Grid, vol. 5,
no. 4, pp. 1696-1703, Jul. 2014.
[5] J. Grainger and S. Cinvalar, “Volt/VAr control on distribution systems
with lateral branches using shunt capacitors and voltage regulators part
I: The overal problem,” IEEE Trans. Power App. Syst., vol. PAS-104,
no. 11, pp. 3278-3283, 1985.
[6] I. Roytelman, B. K. Wee, and R. L. Lugtu, “Volt/VAr control algorithm
for modern distribution management system,” IEEE Trans. Power Syst.,
vol. 10, no. 3, pp. 1454-1460, Aug. 1995.

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSTE.2017.2695195, IEEE
Transactions on Sustainable Energy

10

[7] M. E Baran and Ming-Yung Hsu, “Volt/var control at distribution Carlos F. Sabillon Antunez received the B.Sc. degree in electrical engi-
substations,” IEEE Trans. Power Syst., vol. 14, no. 1, pp. 312-318, neering from the UNAH, Tegucigalpa, Honduras, in 2011, M.Sc. degree in
Aug. 1995. electrical engineering from the São Paulo State University (UNESP), Ilha
[8] V. Borozan, M. E. Baran, and D. Novosel, “Integrated volt/var control Solteira, Brazil, in 2014, and is currently pursuing the Ph.D. degree in
in distribution systems,” In: IEEE Proc. Power Eng. Soc. Winter electrical engineering at UNESP, Ilha Solteira, Brazil. His areas of research are
Meeting, Columbus, OH, vol. 3, pp. 1485-1490, Jan. 2001. the development of methodologies for the optimization, planning and control
[9] B. R. Schainker, “Executive overview: energy storage options for a of electrical power systems.
sustainable energy future,” Power Eng. Soc. General Meeting, vol. 2,
pp. 2309-2314, Jun. 2004.
[10] L. H. Macedo, J. F. Franco, M. J. Rider, and R. Romero “Optimal
operation of distribution networks considering energy storage devices,”
IEEE Trans. Smart Grid, vol. 6, no. 6, pp. 2825-2836, Nov. 2015.
[11] I. Atzeni, L. G. Ordonez, G. Scutari, D. P. Palomar, and J.R. Fonollosa,
“Demand-side management via distributed energy generation and stor-
age optimization,” IEEE Trans. Smart Grid, vol. 4, no. 2, pp. 866-876,
Jun. 2013. Ozy D. Melgar Dominguez received the B.Sc. degree in electrical engineering
[12] S. Lin, Z. He, T. Zang, and Q. Qian, “Impact of plug-in hybrid electric from the UNAH, Tegucigalpa, Honduras, in 2011, M.Sc. degree in electrical
vehicles on distribution systems,” Proc. 2010 Int. Conf. Power Syst. engineering from the São Paulo State University (UNESP), Ilha Solteira,
Technol., pp. 1-5. Brazil, in 2015, and is currently pursuing the Ph.D. degree in electrical
[13] He Yifeng, B. Venkatesh, and Ling Guan, “Optimal scheduling for engineering at UNESP, Ilha Solteira, Brazil. His areas of research are the
charging and discharging of electric vehicles,” IEEE Trans. Smart Grid, development of methodologies for the optimization, planning and control of
vol. 3, no. 3, pp. 1095-1105, Sep. 2012. electrical power systems.
[14] R. Das, K. Thirugnanam, P. Kumar, R. Lavudiya, and M. Singh,
“Mathematical modeling for economic evaluation of electric vehicle
to smart grid interaction,” IEEE Trans. Smart Grid, vol. 5, no. 2, pp.
712-721, Mar. 2014.
[15] A. O’Connell, D. Flynn, and A. Keane, “Rolling multi-period opti-
mization to control electric vehicle charging in distribution networks,”
IEEE Trans. Power Syst., vol. 29, no. 1, pp. 340-348, Jan. 2014.
[16] S. Deilami, A. S. Masoum, P. S. Moses, and M. A. S. Masoum, “Real-
time coordination of plug-in electric vehicle charging in smart grids
John F. Franco (S’11–M’13) received the B.Sc. and M.Sc. degrees in
to minimize power losses and improve voltage profile,” IEEE Trans.
electrical engineering from the Universidad Tecnologica de Pereira, Colombia,
Smart Grid, vol. 2, no. 3, pp. 456–467, Sep. 2011.
in 2004 and 2006, respectively, and the Ph.D. degree in electrical engineering
[17] S. Rahimi, M. Marinelli, and F. Silvestro, “Evaluation of requirements
from the São Paulo State University (UNESP), Ilha Solteira, Brazil, in 2012.
for volt/var control and optimization function in distribution man-
Currently, he is a Professor in the Electrical Engineering Faculty at UNESP,
agement systems,” 2012 IEEE International Energy Conference and
Rosana, Brazil. His areas of research are the development of methodologies
Exhibition (ENERGYCON), Florence, pp. 331-336, 2012.
for the optimization, planning and control of electrical power systems.
[18] V. Dabic, Siew Cheong, J. Peralta, and D. Acebedo,“BC hydro’s
experience on voltage var optimization in distribution system,” 2010
IEEE PES Conf. Expo. Trans. and Dist., pp. 1-7, Apr. 2010.
[19] L. M. Korunovic, S. Sterpu, S. Djokic, K. Yamashita, S. M. Villanueva,
and J. V. Milanovic,“Processing of load parameters based on existing
load models,” Proc. 3rd IEEE Power Energy Soc. Innovative Smart
Grid Technol. Eur., pp. 1-6, 2012 .
[20] “IEEE task force on load representation for dynamic performance,
“Bibliography on load models for power flow and dynamic perfor-
mance simulation,” IEEE Trans. on Power Syst., vol. 10, no. 1, pp. Marina Lavorato (S’07–M’11) received the B.Sc and M.Sc degrees in 2002
523-538, Feb. 1995. and 2004, respectively, from the Federal University of Juiz de Fora, Brazil,
[21] J. F. Franco, M. J. Rider, and R. Romero, “A mixed-integer linear and the Ph.D. degree in 2010 from the University of Campinas (UNICAMP),
programming model for the electric vehicle charging coordination Brazil, all in electrical engineering.
problem in unbalanced electrical distribution systems,” IEEE Trans. Currently, she is a Professor in the Electrical Engineering Faculty at
Smart Grid, vol. 6, no. 5, pp. 2200-2210, Sep. 2015. Pontifical Catholic University of Campinas, Campinas, Brazil. Her areas of
[22] J. D. Melo, E. M. Carreño, and A. Padilha-Feltrin, “Spatial-temporal research are the development of methodologies for the optimization, planning
simulation to estimate the load demand of battery electric vehicles and control of electrical power systems.
charging in small residential areas,” J. of Control, Automation and Elec.
Sys., vol. 25, pp. 470-480, 2014.
[23] K. Qian, C. Zhou, M. Alan, and Y. Yuan, “Modeling of load demand
due to ev battery charging in distribution systems,” IEEE Trans. Power
Sys., vol. 26, no. 2, pp. 802-810, May 2011.
[24] A. Lojowska, D. Kurowicka, G. Papaefthymiou, and L. Van der Sluis,
“Stochastic modeling of power demand due to EVs using copula,” IEEE
Trans. Power Syst., vol. 27, no. 4, pp. 1960-1968, Nov. 2012.
[25] Distribution Test Systems, 33-bus Test System Marcos J. Rider (S’97–M’06–SM’16) received the B.Sc. (Hons.) and P.E.
with Secondary Network. [Online]. Available: degrees from the National University of Engineering, Lima, Peru, in 1999 and
http://www.feis.unesp.br/#!/departamentos/engenharia- 2000, respectively, the M.Sc. degree from the Federal University of Maranhão,
eletrica/pesquisas-e-projetos/lapsee/. Maranhão, Brazil, in 2002; and the Ph.D. degree from the University of Camp-
[26] Tesla Motors - Model Specifications. [Online]. Available: inas (UNICAMP), Campinas, Brazil, in 2006–all in electrical engineering.
http://www.teslamotors.com/models/specs. Currently, he is a Professor in the Department of Systems and Energy
[27] Nissan - Model Specifications. [Online]. Available: at UNICAMP. His areas of research are the development of methodologies
http://www.nissanusa.com/electric-cars/leaf/. for the optimization, planning, and control of electrical power systems, and
[28] R. Fourer, D. M. Gay, and B. W. Kernighan, AMPL: A modeling applications of artificial intelligence in power systems.
language for mathematical programming. CA: Brooks/Cole-Thomson
Learning, Pacific Grove, 2nd Ed., 2003.
[29] “IBM ILOG CPLEX V12.1 User’s Manual for CPLEX,” CPLEX
Division, ILOG Inc., Incline Village, NV, USA, 2009.

1949-3029 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.